this is a try to modularize the original code from:

REINFORCE.js

see also: Original Github Repo

Installation

npm install reinforcenode

Usage

let reinforce = require('reinforcenode');
let Agent = reinforce.DQNAgent;
// create an environment object
let env = {};
env.getNumStates = function() { return 8; }
env.getMaxNumActions = function() { return 4; }

// create the DQN agent
let spec = { alpha: 0.01 } // see full options on DQN page
agent = new Agent(env, spec);

setInterval(function(){ // start the learning loop
  let action = agent.act(s); // s is an array of length 8
  //... execute action in environment and get the reward
  agent.learn(reward); // the agent improves its Q,policy,model, etc. reward is a float
}, 0);

Alpha

for now the module exports one Object

// replaces the former binding to the golbal window object in the browser
module.exports = {
 maxi: maxi,
 samplei: samplei,
 softmax: softmax,
 UTILS: UTILS,
 forwardLSTM: forwardLSTM,
 initLSTM: initLSTM,

 // optimization
 Solver: Solver,
 Graph: Graph,
 //agents for now only one is working
 DQNAgent: DQNAgent

};

notice the UTILS which in turn exports the bundled helper functions:

// a module for all the util functions
module.exports = {
  Mat: Mat,
  RandMat: RandMat,
  assert: assert,
  sig: sig,
  updateMat: updateMat,
  updateNet: updateNet,
  copyNet: copyNet,
  copyMat: copyMat,
  netToJSON: netToJSON,
  netFromJSON: netFromJSON,
  netZeroGrads: netZeroGrads,
  netFlattenGrads: netFlattenGrads,
  getopt: getopt,
  setConst: setConst,
  sampleWeighted: sampleWeighted,
  randi: randi
};